Trois étapes sensibles du travail de géologue de ressources

In mining geology, block models are used for resource estimations. To guarantee quality of the resources estimates in accordance to the Canadian' standard NI43-101, the qualified person relies on a set of formal and informal rules : the best practices. These rules lay down the approaches to be followed for significant subjects such as drilling recommendations, high grade capping and overall validation of recoverable resources. These successive steps are among the most important subjects during the resource estimate process. They can induce multi-millions additional expenses or losses up to hundreds of millions of dollars: these steps are therefore of central importance for projects or mines in operations. However, the vast majority of resource geologists do not use objective and repeatable methods to ensure the effectiveness or quality of their applications of the best practices. In this thesis, we propose new approaches to systematize and rationalize the drilling recommendations and the definitions of capping thresholds. We also present a study to document and reinforce some of the textit best practices regarding the validation of the global recoverable resources using change of support methods according to the variogram type. Any resource estimate begins with a drill hole database and ends with drill holes' recommendation. These recommendations are generally carried out manually and are therefore subjective. Drilling campaigns represent investments up to tens of millions of dollars and are a significant challenge for all mining or exploration companies. To this drilling positioning problem, we propose to rationalize the drilling recommendations with a new operational re- search tool. Flexible, it works for both underground and surface data sets. The algorithm dynamically takes into account the partial coverage of the blocks of existing or recommended drill hole. The algorithm is fast even for large data sets with an optimality gap minimum when compared to linear programming. Applied to an operating mine data, the algorithm allows the optimization of several projects simultaneously. This advantage is particularly useful companies' budgets definition, as it allows a rationalization of the different recommended drilling programs. In application on several projects simultaneously, the impact of the objective function parameters used remains to be established. Flexible and fast, this new tool makes possible to obtain objective drilling recommendations which guarantee maximum coverage according to the allocated budget. Once the drilling campaign is completed, core assays allow the resource estimate to begin. A recurring challenge is then grades variability management and the presence of extreme values. Since a large quantity of metal and therefore potential income is associated with extreme values, their management through capping threshold definitions is a major problem for all resource geologists. The most common practice consists of visually interpret assays histogram to define capping threshold. Thought the use of quality analysis quality control (QaQc) data, we propose a new approach linking sampling quality and extreme values. The capping threshold is selected such that the coefficient of variation of the capped assays is equal to the one estimated forward error-free population (established from the correlation coefficient of duplicates). Using two real data sets from gold and nickel mines, the method demonstrates results consistent with existing methods used in industry. The use of a robust estimator for the correlation coefficient of duplicates represents a possible improvement to the proposed methods. This original approach is objective and repeatable, it enhances the data from QaQc and link the capping threshold to the sampling quality. The proposed method provides a new tool to assist geologists in this critical decision. Once data has been properly capped, the qualified person can proceed with block model interpolation. Most often in the Canadian mining industry a linear estimator is used. The interpolation parameters choice is often done thought the validation of the recoverable re- source estimates by global change of support. The grade-tonnage curves at the desired block size are deduced from the data sets, without an extreme value, providing: the histogram and the variogram. There are four different variograms types commonly used in the mining industry (traditional, correlogram, pairwise and nscore). The geologist-geostatistician must then choose a change of support model. In order to document the consequences of these choices, we studied the impact of the different variograms types on the results of several change of support models. Using simulated data sets, we have corroborated some of the best practices used in the industry. We demonstrated the superiority of the Discrete Gaussian Model is the compared to the indirect log normal methods. The Discrete Gaussian Model is more robust and does not present any bias for all tested grade distributions, as such, it is the recommended model to be used to reduce projects risks. The pairwise relative variogram do not provide a good estimate of block dispersion for change of support models and should therefore be avoided in this context. Existing global change of support models are based on the stationary assumption; however, in most mining projects, this assumption is weak. Therefore, the development of a non-stationary global change of support model represents an interesting direction for future research. This study helps the qualified persons to identify the approaches to be avoided and the ones recommended as best practices to be used for the validation of the resource estimates by global change of support methods. In this thesis, we proposed several original solutions as improvements to the Canadian' resource estimation best practices. Using both simulated and real data sets from projects and operating mines, we assessed the proposed algorithms and methods. The results of our studies corroborate current practices while providing robust and objective improvements. The use and respect of the methods developed in this thesis improve the best practices, reducing the risk associated with resource estimates and maximizing the value of exploration projects and operating mines.